Critical Parted Question

[Jeremy Moles]

I have a quick question about parted. I doubt its a bug in parted
itself; rather, it is more likely that I am simply using it
incorrectly. :)

Okay, here goes:

I use parted in a large bash script to automate a lot of tedious (and
difficult using any tool that doesn't let me tell it a partition size in
megabytes) partition manipulation. The process is as follows:

1. I receive a machine with a partition table containing Windows and a
special "Recovery Partition."

2. I shrink the windows partition using ntfsresize and then go in using
fdisk to recreate the partition layout.

3. At this point, the recovery partition still works no matter how I get
in to it (via the BIOS option or using a GRUB CD to boot into it).

4. Another step comes through later and uses parted to fill in the
remaining space with useful stuff; HOWEVER, for some reason, parted
changes the original disk geometry parameters. For example: the disk
originally had 240 heads, but after parted "touches" it, this gets
changed to a value like 16.

5. At this point, Windows (hda1) will boot just fine. The recovery
partition, (hda2), is no longer accessible using the BIOS's recovery
mechanism--which, for IBMs, is the "Access IBM" button. Additionally,
any attempt to "use" it results in the standard "Blue Screen of Death."

Unfortunately, this is irreversible... no matter what I do, I cannot
recreate the partition (using the original geometry constraints) to make
the recovery work again.

My question is: why is parted modifying these values? Is there an
invocation that says: "don't touch the number of heads, etc." As it
stands, parted seems to do what it wants to the geometry parameters
(though, I'm sure smarter people than me wrote parted and have a good
reason WHY they're doing it :) ), but it's seriously breaking the
recovery partition.

Anyone have any ideas?